| name | add-news |
| description | Add, update, translate, or correct a lab news item in Lab Index. Use for major model announcements, funding, leadership changes, partnerships, policy news, and non-English article summaries, including checking whether the news also requires a structured research output. |
Add a News Item to a Lab
Follow these steps to add a news article to a lab's news section.
1. Prepare the Entry
Each news item requires:
- title: Article headline (in English, even for non-English sources)
- url: Link to the article
- source: Publication name (e.g., Bloomberg, SCMP, The Information)
- date: Publication date in
YYYY-MM-DD format
2. Add to Lab YAML
Edit data/labs/{slug}.yaml and add to the news array:
news:
- title: "Article Headline Here"
url: https://www.example.com/article
source: Publication Name
date: 2026-04-02
Ordering
Add new items in reverse chronological order (newest first).
3. Non-English Articles
For articles in Chinese or other non-English languages:
- Create an English summary as a rendered Astro page at
src/pages/articles/{slug}.astro
- Use the site Layout component and HTML for proper rendering
- Include a link to the original article at the top of the summary
- Add a single news entry pointing to the English summary:
- title: "Article Title (English summary)"
url: /articles/summary-slug
source: Original Publication
date: 2026-03-28
News URLs can be:
- Absolute URLs (
https://...) for external articles
- Relative paths (
/articles/...) for locally hosted summaries
4. When to Add News
Add news items for:
- Major model releases or announcements
- Organizational changes (restructuring, key hires/departures)
- Funding rounds or IPO filings
- Strategic partnerships or policy changes
- Significant press coverage that provides context about the lab
Do NOT add news for:
- Routine paper publications (these are outputs, not news)
- Minor product updates
- Social media posts
News + output: when the news references a research artifact, file BOTH
A press release announcing a model, dataset, or library is also a flag that an output entry should exist. Filing only the news leaves the canonical artifact untracked (no intelligence_index, no openness_index, no listing on the lab's outputs surface). Real misses caught by this rule:
- TII Falcon Perception — news entry filed April 1 from the TII launch post; the companion arXiv 2603.27365 paper, HuggingFace models (Falcon-Perception, Falcon-Perception-300M, Falcon-OCR), and PBench benchmark sat untracked for ~2 months because we never ran the output-probe path on the same announcement.
Whenever the news headline contains words like "launches," "releases," "unveils," "announces," or names a specific model / paper / dataset, run the same external-link probes the add-output skill specifies:
- Search arXiv for a companion paper from the same lab within ±14 days of the news date.
- Hit the lab's HuggingFace org via
curl -sL 'https://huggingface.co/api/models?author=<org>&sort=lastModified&direction=-1&limit=30' (plus the parallel /api/datasets query) and look for uploads matching the announcement.
- Check the lab's GitHub for a companion repo (often named after the model).
- Check the lab's own blog and HuggingFace blog for a technical post with benchmarks.
If any of those turn up an artifact, add an output entry per .agents/skills/add-output/SKILL.md in the same pass. The news entry stays as-is (it's the human-context wrapper); the output is the structured artifact record that lets the rest of the index work (intelligence_index, openness_index, benchmarks, model details, AAII/AAOI fetch scripts, etc.).
5. Finding News — Priority Sources
When searching for news about a lab, check sources in this order. Prefer higher-tier sources when multiple outlets cover the same story.
Tier 1 — Check first (highest signal):
- Lab's own blog/newsroom — always the first and most authoritative source
- Bloomberg — funding rounds, valuations, strategic pivots, China AI
- TechCrunch — startup fundraising and product launches
- CNBC — major corporate AI milestones
- The Information — exclusive scoops on strategy, unreported fundraising
Tier 2 — Check for regional/specialist coverage:
6. ChinaTalk — deep analysis of Chinese labs
7. SCMP — broadest English-language Chinese tech/AI
8. VentureBeat — enterprise AI, open-weight model launches
9. Caixin — Chinese business investigative journalism
10. LatePost (晚点) — Chinese investigative tech (often breaks stories first)
11. Reuters — global AI corporate news, regulation
12. 36Kr (36氪) — premier Chinese startup news
13. QbitAI (量子位) — leading Chinese AI news aggregator at qbitai.com; dense daily coverage of Chinese-lab model releases, funding, and papers. Aggregator — cross-check upstream before citing. Translate via src/pages/articles/ (see step 3) when adding as a news entry.
Tier 3 — Region or event-triggered:
13. Nikkei Asia — Japan/Korea lab coverage
14. Korea Times — Korean labs (Kakao, Naver, SKT, LG, Upstage)
15. GeekWire — Pacific NW (AI2, Amazon)
16. Calcalist — Israeli tech (AI21)
17. Pandaily — China AI in English
18. Wired / Ars Technica / The Verge — major launch coverage, longer analysis
19. Latent Space / Interconnects — community deep-dives, technical context
6. Validate
npm run validate
npm run build
